Welcome to NCA-AIIO Labs: Schedule Your First GPU Pod

Your isolated Kubernetes cluster

When you started this lab, the platform provisioned a fresh virtual Kubernetes cluster (vCluster) just for you. You have full cluster-admin inside it — you can create namespaces, ClusterRoles, ResourceQuotas, the works. The cluster sees one Node (the host machine) advertising fake NVIDIA GPUs.

Step 1a — kick the tires

Run these in your terminal to get oriented:

kubectl get nodes -o wide
kubectl describe node | grep -A2 'Capacity:\|nvidia.com/gpu'
kubectl get runtimeclass

You should see:

• One node with status Ready
• Capacity and Allocatable lines mentioning nvidia.com/gpu: 2
• A RuntimeClass named nvidia (handler: runc — the mock GPU runtime)

Step 1b — schedule a GPU Pod

Now open the file gpu-test.yaml. It's a Pod manifest with a few # TODO: markers. Fill them in so the Pod:

• Uses the nvidia RuntimeClass
• Requests one GPU via resources.limits
• Runs nvidia-smi and exits

Then apply it and inspect the output:

kubectl apply -f gpu-test.yaml
kubectl wait --for=condition=Ready pod/gpu-test --timeout=30s || true
sleep 5
kubectl logs gpu-test

When nvidia-smi reports a Tesla K80 with realistic-looking utilization numbers, you've verified the scheduling chain end-to-end:

1. Your manifest passed admission (PSS=baseline + Kyverno's policies)
2. The scheduler bound the Pod to a node with nvidia.com/gpu capacity
3. The fake GPU runtime mounted the mock nvidia-smi into your container
4. The binary contacted the topology server to render plausible data

That's the same chain a real NVIDIA GPU Operator deployment uses — only the bottom layer (real driver vs mock) differs.

NCA-AIIO exam note: You should be able to read a nvidia-smi output and extract: driver version, CUDA version, GPU memory total/used, GPU-Util, and identify when a Process is consuming the GPU vs idle. This lab gives you a baseline shape to refer to.

apiVersion: v1
kind: Pod
metadata:
  name: gpu-test
spec:
  # TODO: opt into the nvidia RuntimeClass
  runtimeClassName: ______
  restartPolicy: Never
  containers:
    - name: gpu
      image: docker.io/library/ubuntu:22.04
      command: ["sh", "-c", "sleep 3; nvidia-smi; exit 0"]
      resources:
        limits:
          # TODO: request one fake GPU
          nvidia.com/gpu: ______